Presentation on theme: "Agreement Assessment of Visual Interpretation and Digital Classification for Mapping Urban Landscape Heterogeneity Weiqi Zhou, Kirsten Schwarz, Mary Cadenasso."— Presentation transcript:
Agreement Assessment of Visual Interpretation and Digital Classification for Mapping Urban Landscape Heterogeneity Weiqi Zhou, Kirsten Schwarz, Mary Cadenasso 2008 BES Annual Meeting
Motivations (1) Visual interpretation of remotely sensed images is extensively used for urban analysis. – Patch mapping – Patch classification: Within-patch composition estimation. However, few studies evaluate accuracy of the within-patch composition estimations, particularly in urban settings.
Motivations (2) Digital classification of high resolution image – Object-based classification greatly increases the accuracy of digital classification – Digital classification as reference data
Visual interpretation Cadenasso et al., 2007
Research Questions What is the relative agreement of percent cover estimation between the two methods? What are the spatial patterns of the patches with large disagreement?
# of HERCULES patches: 2250 Degree of Disagreement Digital Classification Visual Interpretation
What is the relative agreement between the two classification methods? Standard Procedure: the strictest agreement assessment method Plus-one Method: A modification of the standard procedure, and accepts plus or minus one class of the actual class as agreement Fuzzy Set Theory: Create fuzzy rules to account for fuzzy class boundaries.
Observations from Agreement Assessment Largest agreement when patches were dominated by one type of cover Largest disagreement for cover ranges %, 35-75% Vegetation: underestimated when cover <35% Pavement: Underestimated Buildings: Overestimated.
Do the patches with large disagreement cluster spatially in the watershed? Getis-Ord General G index: test whether the patches with large disagreement tend to cluster spatially Anselin local Moran’s I index: detect spatial clusters (i.e., hot spots) of large disagreement within the watershed.
Pattern Analysis: General G Index Except for fine vegetation, patches with large disagreement clustered spatially (p<0.05). Landscape FeaturesObserver Gp-value CV FV Bare soil <0.01 Pavement <0.01 Buildings <0.01
CV FV Bare Soil Pave Building
Next step: How does patch heterogeneity affect the degree of disagreement? Patch complexity metrics: e.g. patch size, shape, etc. Within-patch heterogeneity metrics – Patch composition – Within-patch configuration
Acknowledgements This research was funded by the National Science Foundation LTER program (grant DEB ) and biocomplexity program (grant BCE ). Many thanks to a lot of BES people and colleagues in the Cadenasso lab.